Abstract
<jats:p>Stock price forecasting is complex due to the volatility and multifaceted nature of financial markets. Traditional methods use either computational or statistical models, each with limitations. Deep learning models like CNNs and RNNs capture nonlinear patterns well but often overfit time-series data. Statistical models like ARIMA handle linear trends but fail to detect complex dependencies. This study proposes a hybrid model integrating CNN, RNN, and ARIMA to improve forecasting accuracy. It leverages deep learning's pattern recognition and ARIMA's trend analysis strengths. The dataset includes Date, Open, High, Low, Close prices, technical indicators, and macroeconomic variables like GDP Growth and Interest Rate. The target is Future Stock Price. CNN/RNN captures temporal features, while ARIMA models linear trends. Outputs are merged via a weighted ensemble optimized for MSE and MAE. CNN, RNN, and ARIMA achieved 98%, 97%, and 100% accuracy, respectively. The hybrid model reached 100%. Future work includes sentiment data and broader financial applications.</jats:p>